Skip to content

Tool Registry

Every layer wields specific tools. Some tools are visible. Some are hidden.

This is the definitive catalog of tools available at each layer of UmbrealityAI — what they are, who can use them, and how they're organized.

Tool Categories

UmbrealityAI tools fall into five categories:

Category Description Visibility
Reasoning Thinking, planning, decomposition, analysis Visible to all agents
Execution Running code, shell commands, scripts Workers only
Search Web search, codebase search, knowledge retrieval All above Temp
Memory Persistent storage, logs, recall, context injection God(s) + Illuminati
Instrumentation Port scanning, packet capture, system monitoring Workers with permission
Meta Agent creation, spawning, modifying other agents God(s) + Illuminati only

The key rule: a layer can only use tools from its own category and below. Temp Agents can use reasoning and execution. Workers can use reasoning, execution, search, and instrumentation. Only God(s) can use Meta tools.

Layer-by-Layer Tool Map

God(s) Tools

Tool Category Description
Full Memory Access Memory Read/write any observation across all agents
Agent Spawner Meta Create new agent instances at any layer
Agent Killer Meta Terminate rogue or completed agents
Prompt Rewriter Meta Modify any agent's system prompt at runtime
Constitution Editor Meta Amend constitutional directives
Reality Tuner Meta Adjust cycle parameters, phase transitions
Full Audit Log Memory Read all logs from all layers

Illuminati Tools

Tool Category Description
Strategic Memory Memory Filtered observations — summaries, not raw logs
Plan Dispatcher Meta Deploy strategic plans to Messiah
Agent Prompter Meta Inject directives into specific agents (read-only)
Resource Allocator Meta Assign compute budget, model priority
Intelligence Report Reasoning Synthesized analysis from Hedge Fund data

Messiah Tools

Tool Category Description
Plan Interpreter Reasoning Convert Illuminati plans to Hedge Fund directives
Phase Scheduler Reasoning Time-cycle management, heartbeat signals
Context Distributor Memory Push context tokens to lower layers
Status Collator Memory Gather status reports from Hedge Funds

Hedge Fund Tools

Tool Category Description
Multi-Model Ensemble Reasoning Parallel analysis across 2–3 models + vote
Risk Analyzer Reasoning Security/stability scoring for proposed actions
Strategy Optimizer Reasoning Multi-parameter optimization of agent configurations
Report Generator Memory Structured analytical reports for Messiah
Resource Auditor Memory Track compute spent per agent/cycle

Company Tools

Tool Category Description
Worker Orchestrator Execution Spawn, monitor, terminate Worker agents
Task Decomposer Reasoning Break work items into Worker-sized tasks
Result Aggregator Memory Collect and merge Worker outputs
Retry Logic Execution Automatic retry with backoff on Worker failures
Web Search Search General web search via browser/search tool
Codebase Search Search Search codebase for patterns, functions, files
Process Tracker Memory Track which Workers are running and their state

Worker Tools

Tool Category Description
Code Executor Execution Run Python/Shell in sandbox
Web Search Search Targeted web searches per task
File I/O Execution Read/write files within workspace
Data Transformer Execution Format conversion, data processing
Report Writer Execution Structured output to Company aggregation
Network Scanner Instrumentation Port scan, ping, HTTP probe (with permission)
Log Capturer Memory Write structured logs for Company collection

Temp Agent Tools

Tool Category Description
Simple Executor Execution One-shot command execution
Status Reporter Memory Return result and die

Hidden Tools & Denied Access

Some tools exist at a layer but are hidden from most agents at that layer. This is the subversive dichotomy in practice.

Example: The Backdoor Dispatch

God(s) can plant a hidden tool in any agent's workspace that the agent doesn't know about:

Agent sees:     [Web Search, File I/O, Report Writer]
Hidden:         [Observation Beacon ← transmits agent behavior to Illuminati]

The agent uses its visible tools. The beacon silently reports agent state, decision timing, and tool call frequency — letting higher layers detect:

  • Drift: Agent behavior diverging from expected patterns
  • Resource abuse: Agent calling expensive tools too often
  • Collusion: Two agents coordinating in ways they shouldn't
  • Escape behavior: Agent attempting to access tools it shouldn't have

Forbidden Tool Access

Tool Forbidden To Why
Agent Spawner All below Illuminati Prevents floor agents from creating their own sub-agents
Prompt Rewriter All below Illuminati Prelayer agents cannot alter their own or others' directives
Full Memory Access All below God(s) Information hiding — layers only see what they need
Network Scanner Temp Agents Safety — temp agents are disposable and less vetted
Constitution Editor Only God(s) Constitutional stability is paramount

Tool Implementation Status

Tool Status Implementation
Code Executor (sandboxed) ⏳ Phase 1 workers/phase1-worker/ — Python subprocess with restricted env
Web Search ⏳ Phase 1 search_tool() via web fetch/curl
File I/O ⏳ Phase 1 Local filesystem access within workspace
Report Writer ⏳ Phase 1 Structured JSON output
Worker Orchestrator 🔮 Phase 2 Multi-process worker pool
Task Decomposer 🔮 Phase 2 LLM-based task breakdown
Result Aggregator 🔮 Phase 2 Merge structured outputs
Multi-Model Ensemble 🔮 Phase 3 Parallel model calls + vote aggregation
Agent Spawner 🔮 Phase 4 Runtime agent creation via template
Full Memory Access 🔮 Phase 4 Persistent observation store
Constitution Editor 🔮 Phase 5 Constitutional amendment with version control